能量图空间分解法和DSW相结合的步态识别

Gait recognition method based on space mass energy image and Dynamic Space Warping

提出一种能量图空间分解法和DSW(动态空间规整法)相结合的步态识别算法;提取二值化步态轮廓序列,通过计算差分图像获取步态序列的活动能量图,在空间域进行行列分解;通过加窗和傅里叶变化等对分解向量进行频谱分析,并对频谱曲线求极值来构建步态特征向量;采用DSW算法,进行空间领域的步态分类识别。使用中科院自动化研究所的Dataset B数据库进行了充足的实验,最终的平均识别率达到了93%,并对衣服背包具有较强的鲁棒性,与其他相关文献的对比显示该算法具有更好的识别性能。

This paper proposes a gait recognition algorithm based on mass energy image and spectrum analysis. The method includes following steps： firstly, two-value silhouette sequences extraction is performed for each subject; active energy image is calculated by adding up the difference of each image in a sequence; the row mass vector and column mass vector are extracted respectively. Secondly, the space mass energy image is spectrum analysed by windowed and Fourier trans- form techniques, combining the extreme-value computing method to construct gait feature vector. It uses dynamic space warping for classification under space mass energy image and spectrum analysis. Enough experiments are performed on CASIA Dataset B. The final average recognition rate reaches 93%, and with a robustness of backpacks and clothes. The method has been demonstrated higher recognition rate than other methods.